Fastest models

Provider-observed p50 output-throughput ordering. The API does not expose tok/s and places unmeasured models last.

1. 1
   Morph: Morph V3 Largemorph/morph-v3-large
   #1
2. 2
   Relace: Relace Apply 3relace/relace-apply-3
   #2
3. 3
   Morph: Morph V3 Fastmorph/morph-v3-fast
   #3
4. 4
   OpenAI: gpt-oss-safeguard-20bopenai/gpt-oss-safeguard-20b
   #4
5. 5
   OpenAI: gpt-oss-120bopenai/gpt-oss-120b
   #5
6. 6
   SpaceXAI: Grok 4.20 Multi-Agentx-ai/grok-4.20-multi-agent
   #6
7. 7
   NVIDIA: Nemotron 3.5 Lightningnvidia/nemotron-3.5-lightning
   #7
8. 8
   Inference.net: Schematron V2 Smallinference-net/schematron-v2-small
   #8

Provider p50 time-to-first-token ordering with unmeasured models placed last, not an intrinsic fixed model speed.

1. 1
   IBM: Granite 4.2 8Bibm-granite/granite-4.2-8b
   #1
2. 2
   Baseten: Joey 1baseten/joey-1
   #2
3. 3
   TypeSafe: Jev 1.13typesafe/jev-1.13
   #3
4. 4
   VoyageAI by MongoDB: voyage-4-largemongodb/voyage-4-large
   #4
5. 5
   Together: Tev1 4B Experimentaltogethercomputer/tev1-4b-experimental
   #5
6. 6
   Baseten: Elephant 1baseten/elephant-1
   #6
7. 7
   TypeSafe: Jev Latest~typesafe/jev-latest
   #7
8. 8
   TypeSafe: Jev 1.14typesafe/jev-1.14
   #8